JARVIS Research OS

AI-Powered Research Operating System for Systematic Literature Reviews

JARVIS Research OS is a local-first, AI-powered research assistant that automates systematic literature reviews with evidence grading, citation network analysis, and multi-agent orchestration.

Release Status

• Current version: v2.0.0 (2026-03-06)
• CLI Commands: 22

Features

Core Pipeline

• Multi-Source Search: PubMed, Semantic Scholar, OpenAlex, arXiv, Crossref
• Evidence Grading: CEBM evidence level classification (1a-5)
• Paper Scoring: Comprehensive quality scoring with multiple signals
• LLM Japanese Summary: Gemini-powered structured summaries
• Deduplication: DOI/title fuzzy matching with RapidFuzz

Analysis & Discovery

• Citation Network: PageRank hub detection, Mermaid/GraphML visualization (D5)
• Citation Stance: Support/Contrast/Mention classification
• Contradiction Detection: Negation, antonym, and quantitative analysis
• Semantic Search: ChromaDB persistent vector search (D3)
• Deep Research: Autonomous multi-query research with Codex/Copilot/Gemini fallback (D4)
• Active Learning: Efficient screening with uncertainty sampling

AI Infrastructure

• LiteLLM: Unified multi-provider LLM access (Gemini/OpenAI/DeepSeek) (D1)
• LangGraph Orchestrator: 6-agent pipeline with conditional retry loops (D4)
• PydanticAI + Instructor: Structured LLM output validation (D1)
• LightRAG: Graph-based RAG with entity/relation extraction (D3)
• PDF-to-Markdown: PyMuPDF4LLM full-text extraction (D3)

Export & Integration

• Obsidian Notes: Auto-generated research notes in Obsidian Vault
• Zotero Sync: Reference management integration
• PRISMA Diagrams: PRISMA 2020 flow diagram generation (Mermaid)
• BibTeX Export: Standard bibliography format
• Google Drive: H: drive storage with local fallback (D5)

Ecosystem

• MCP Hub: 5 servers, 15 tools for external API access
• Skills System: Task-specific workflows via SKILL.md
• Streamlit Dashboard: 5-page web UI (Overview/Search/Citation/ChromaDB/Storage) (D6)

Agent-Web (Chat UI + Research Pipeline)

JARVIS includes a browser-based agent-web/ app that unifies chat, research pipeline automation, and local data management on top of Express + SQLite.

Features

• Real-time SSE Chat: streaming responses with inline activity timeline
• Daily Digest: scheduled multi-keyword paper collection, Gemini summary, SQLite persistence, and Obsidian export
• Papers CRUD: local paper repository APIs for create/read/update/delete workflows
• Pipeline Execution: end-to-end 7-step research pipeline with SSE progress events
• PDF Archive: automatic PDF download and archive flow from collected papers
• Zotero Sync: Zotero Web API integration for bibliography synchronization
• ChromaDB Bridge: Python bridge for semantic indexing and vector retrieval
• H-2 Session Memory: recent conversation context injection per session
• H-3 Persistent Memory: cross-session memory via facts and user_preferences
• Memory API: /api/memory/* endpoints for facts, preferences, and context
• Test Coverage: 47 tests passed (latest verification: 2026-03-06)

Tech Stack

• Backend: Express v5, better-sqlite3, SSE, Copilot proxy + model fallbacks
• Frontend: Vanilla JS SPA, dark/gold UI, animated timeline interactions
• Data Layer: chat/session/papers/memory SQLite tables + bridge modules

Quick Start (Agent-Web)

# Terminal A: Start Copilot API proxy
cd agent-web
npx copilot-api@latest start --port 4141

# Terminal B: Start Agent-Web server
cd agent-web
npm run dev
# => http://localhost:3000

# Terminal C: Run tests
cd agent-web
npm test

Agent-Web Architecture

agent-web/
  src/
    server.js                  # Express entry point
    routes/
      chat.js                  # /api/chat/stream (SSE)
      pipeline.js              # /api/pipeline/*
      sessions.js              # /api/sessions CRUD
      papers.js                # /api/papers CRUD
      memory.js                # /api/memory/*
      models.js                # /api/models
      skills.js                # /api/skills
    db/
      database.js              # SQLite bootstrap
      papers-repository.js     # papers CRUD repository
      memory-store.js          # H-3 persistent memory store
      chroma-bridge.js         # ChromaDB Python subprocess bridge
    skills/
      daily-digest.js          # digest orchestrator
      digest-to-obsidian.js    # Obsidian exporter
      pdf-archiver.js          # PDF archive helper
      zotero-sync.js           # Zotero sync helper
  public/
    index.html
    css/styles.css
    js/app.js
  tests/
    *.test.js

Quick Start

# Clone
git clone https://github.com/kaneko-ai/jarvis-ml-pipeline.git
cd jarvis-ml-pipeline

# Setup (Windows)
python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -e .

# Verify
python -m jarvis_cli --help
Copy
CLI Commands (22)
#	Command	Description
1	run	Run full pipeline with goal
2	search	Search papers across multiple sources
3	merge	Merge and deduplicate JSON results
4	note	Generate LLM research note
5	citation	Fetch citation counts
6	citation-stance	Classify citation relationships
7	prisma	Generate PRISMA 2020 flow diagram
8	evidence	Grade CEBM evidence level
9	score	Calculate paper quality scores
10	screen	Active learning paper screening
11	browse	Fetch URL and extract metadata
12	skills	Manage task-specific workflows
13	mcp	MCP Hub server/tool management
14	orchestrate	LangGraph multi-agent orchestrator
15	obsidian-export	Export papers to Obsidian Vault
16	semantic-search	ChromaDB persistent semantic search
17	contradict	Detect contradictions between papers
18	zotero-sync	Sync papers to Zotero library
19	pdf-extract	Convert PDF to Markdown
20	deep-research	Autonomous deep research mode
21	citation-graph	Build citation network visualization
22	pipeline	Run full pipeline with all options
Usage Examples
Copy# Search papers
jarvis search "PD-1 immunotherapy" --max 10 --sources pubmed,s2

# Full pipeline with Obsidian export
jarvis pipeline "spermidine autophagy aging" --max 20 --obsidian --zotero --prisma

# Multi-agent orchestration
jarvis orchestrate run --goal "immunosenescence mechanisms" --max 5

# Deep research
jarvis deep-research "PD-1 resistance mechanisms" --max-sources 20

# Citation network
jarvis citation-graph results.json --obsidian

# Semantic search (ChromaDB)
jarvis semantic-search "autophagy aging" --top 10

# Evidence grading
jarvis evidence papers.json --use-llm

# Streamlit Dashboard
streamlit run jarvis_web/streamlit_app.py
Architecture
jarvis-ml-pipeline/
├── jarvis_cli/          # 22 CLI command handlers
├── jarvis_core/
│   ├── embeddings/      # ChromaDB PaperStore, sentence-transformers
│   ├── evidence.py      # CEBM evidence grading
│   ├── llm/             # LiteLLM client, Pydantic models
│   ├── mcp/             # MCP Hub (5 servers, 15 tools)
│   ├── pdf/             # PDF-to-Markdown (PyMuPDF4LLM)
│   ├── rag/             # LightRAG, CitationGraph, citation enricher
│   ├── skills/          # Skills engine
│   ├── sources/         # Unified multi-source search
│   └── storage_utils.py # H: drive / local fallback paths
├── jarvis_web/          # Streamlit dashboard
├── tests/               # pytest suite (50+ D6 tests)
├── config.yaml          # Project configuration
└── .env                 # API keys (gitignored)
Requirements
Python 3.11+
Windows 11 (primary), Linux/macOS (untested)
Gemini API key (free tier: 15 RPM)
Optional: Codex CLI (ChatGPT Plus), Copilot CLI (Education Pro)
Development
Copy# Run tests
python -m pytest tests/ -v --timeout=30

# Run D6 tests only
python -m pytest tests/test_imports.py tests/test_citation_graph.py tests/test_storage_utils.py -v

# Check CLI
python -m jarvis_cli --help
Version History
Version	Date	Highlights
v2.0.0	2026-03-06	Agent-Web: Daily Digest, Papers CRUD, PDF archive, Zotero sync, ChromaDB bridge, H-2/H-3 memory, Memory API, 47 tests
v1.3.0	2026-03-04	D5: Obsidian/storage H: migration, citation network, 22 CLI commands
v1.2.0	2026-03-03	D3: ChromaDB, LightRAG, PDF-to-Markdown
v1.1.0	2026-03-03	D2: Jina Reader, MCP handlers, browse enhancement
v1.0.0	2026-03-02	Initial release: 21 CLI commands, full SLR pipeline
License
MIT License - see LICENSE for details.

Handover
See HANDOVER_v16.md for complete development context.